Astronomy & Astrophysics manuscript no. Wasp-103_v5 c©ESO 2017August 22, 2017

Signs of strong Na and K absorption in the transmission spectrumof WASP-103b

M. Lendl1, 2, P. E. Cubillos1, J. Hagelberg3, 4, A. Müller2, I. Juvan1, 5, and L. Fossati1

1 Space Research Institute, Austrian Academy of Sciences, Schmiedlstr. 6, 8042 Graz, Austriae-mail: monika.lendl@oeaw.ac.at

2 Max Planck Institute for Astronomy, Königstuhl 17, 69117 Heidelberg, Germany3 Institut de Planétologie et d’Astrophysique de Grenoble, Université Grenoble Alpes, CS 40700, 38058 Grenoble Cédex 9, France4 University of Hawaii, 2680 Woodlawn Dr., Honolulu HI 96822, USA5 Institut für Geophysik, Astrophysik und Meteorologie, Karl-Franzens-Universität, Universitätsplatz 5, 8010 Graz, Austria

ABSTRACT

Context. Transmission spectroscopy has become a prominent tool for characterizing the atmospheric properties on close-in transitingplanets. Recent observations have revealed a remarkable diversity in exoplanet spectra, which show absorption signatures of Na, K andH2O, in some cases partially or fully attenuated by atmospheric aerosols. Aerosols (clouds and hazes) themselves have been detectedin the transmission spectra of several planets thanks to wavelength-dependent slopes caused by the particles’ scattering properties.Aims. We present an optical 550 – 960 nm transmission spectrum of the extremely irradiated hot Jupiter WASP-103b, one of thehottest (2500 K) and most massive (1.5 MJ) planets yet to be studied with this technique. WASP-103b orbits its star at a separation ofless than 1.2 times the Roche limit and is predicted to be strongly tidally distorted.Methods. We have used Gemini/GMOS to obtain multi-object spectroscopy throughout three transits of WASP-103b. We used relativespectrophotometry and bin sizes between 20 and 2 nm to infer the planet’s transmission spectrum.Results. We find that WASP-103b shows increased absorption in the cores of the alkali (Na, K) line features. We do not confirm thepresence of any strong scattering slope as previously suggested, pointing towards a clear atmosphere for the highly irradiated, massiveexoplanet WASP-103b. We constrain the upper boundary of any potential cloud deck to reside at pressure levels above 0.01 bar. Thisfinding is in line with previous studies on cloud occurrence on exoplanets which find that clouds dominate the transmission spectraof cool, low surface gravity planets while hot, high surface gravity planets are either cloud-free, or possess clouds located below thealtitudes probed by transmission spectra.

Key words. planetary systems – stars: individual: WASP-103 – techniques: spectroscopic

1. Introduction

Due to their high temperatures and extended atmospheres, hotJupiters (i.e, gas giants at orbital periods of a few days) areideal laboratories for studying the properties of exoplanet atmo-spheres. Transmission spectroscopy probes the planetary atmo-sphere’s composition and structure by measuring the spectrallyresolved absorption during transit. A growing sample of hot-Jupiter transmission spectra from ground and space is reveal-ing a remarkable diversity in spectral signatures. Some planetsshow prominent absorption signatures of alkali (Na, K), and wa-ter (e.g., Charbonneau et al. 2002, Deming et al. 2013, Nikolovet al. 2016, Sedaghati et al. 2016). Many objects however showattenuated features (e.g., Chen et al. 2017, Sing et al. 2015,2016), and in some the elemental or molecular absorptions areentirely absent at low spectral resolution (Pont et al. 2008, Lendlet al. 2016). Several planets show pronounced slopes of increas-ing absorption towards shorter wavelengths, as can be explainedby scattering on aerosols with small particle sizes (clouds andhazes) lifted to high altitudes in the planetary atmospheres. Thelevel of observed attenuation of the absorption features dependson the particles’ abundance, size and atmospheric altitude. Theobserved slopes stem from scattering on small aerosoles andthe slope extension and amplitude depends on the atmospheric

temperature and the size distribution of the scattering particles(Lecavelier Des Etangs et al. 2008a).

The relation between planetary characteristics and the oc-currence and properties of aerosols is yet to be definitely un-derstood. Stevenson (2016) and Heng (2016) propose that plan-ets with hot atmospheres posses less pronounced cloud features,Stevenson (2016) also suggests that clouds are more common forlow surface gravity planets.

In this work, we present the transmission spectrum of one ofthe most irradiated transiting hot Jupiters known to date, WASP-103b. WASP-103b stands out due to its short orbital period of0.9 d, which is far below the peak of the hot Jupiter perioddistribution of about three days (e.g., Udry et al. 2003). Dueto tidal dissipation of orbital energy, most hot Jupiters are suf-fering from orbital decay and are spiraling into their host stars(e.g., Matsumura et al. 2010, Patra et al. 2017). WASP-103b isa hot Jupiter in its final stages of orbital decay. Currently, theplanetary orbital separation from the host star amounts to only1.16 times its Roche limit, leading to large tidal distortion of theplanet (Budaj 2011, Leconte et al. 2011). This fact, paired withthe planet’s extremely high stellar irradiation makes this objectprone to enhanced mass loss (Lammer et al. 2003, Vidal-Madjaret al. 2003).

Article number, page 1 of 12

arX

iv:1

708.

0573

7v1

[as

tro-

ph.E

P] 1

8 A

ug 2

017

A&A proofs: manuscript no. Wasp-103_v5

We describe our GMOS observations in Sect. 2, and detailour analysis approach, and the correction of our measurementsfor contamination from a nearby star in Sect. 3. In Sect. 4, wepresent and discuss our results for WASP-103b, before conclud-ing in Sect. 5.

2. Observations and data reduction

2.1. GMOS-North observations

We observed WASP-103 throughout three full transits withGMOS at Gemini North during the nights of 27 June 2015, 10July 2015, and 03 May 2016 (all UT, program numbers GN-2015A-Q-63 and GN-2016A-Q-26, PI: Hagelberg). All obser-vations were carried out in multi-object spectroscopy mode us-ing the R400 grism and order separation filter OG515. We usedcustom-cut masks to place slits on the target and two referencestars. The slit width was set to 10 arcsec for all stars to avoidslit losses due to guiding inaccuracies or seeing variations, andthe slit length was set to 40 arcsec for the target and the brighterreference to allow for a good sky determination. For the fainterreference, the slit needed to be kept shorter (25 arcsec) to avoidcontamination from a nearby star.

Spectral coverage of the target and the brighter reference staris between 550 and 980 nm (top and middle spectra in Fig. 1),while spectra of the fainter reference star (bottom spectrum inin Fig. 1) are slightly offset and cover 530 to 960 nm. Two am-plifiers were used to read each of the three CCDs in 1x1 bin-ning and “Slow read" mode. We used custom regions of interest(ROI) to window the detectors and read only the detector areascontaining data. Sky conditions were clear throughout all threeobservations, however, the observations on 10 July 2015 wereinterrupted for approx. 22 minutes due to high ambient humid-ity. We used exposure times of 240 s on 27 June 2015 and 10July 2015, and exposure times of 180 s on 03 May 2016.

2.2. Spectral extraction and wavelength calibration

We carried out the standard calibrations using the GMOS IRAFroutines, and also used these routines to calculate an initial wave-length calibration based on (CuAr) arc frames. The spectra havea pixel scale of 0.68 Å per pixel in the dispersion direction, anda spectral resolution of R ∼ 1200. Cosmic rays were removedusing the LA-cosmic routines (van Dokkum 2001). The result-ing 2D spectra were analyzed using custom-built IDL routines.To correct the spectra for the sky background, we selected tworegions well above and below the stellar spectrum in the spatialdirection. For each spatial pixel, we fitted a linear trend to theflux values in these regions, determined the sky below the stel-lar spectrum via interpolation, and subtracted it. This procedureserves to remove any background trends of instrumental or phys-ical origin, however for the data described here, no appreciabletrends were seen across the sky background in the spatial direc-tion. After sky removal, the spectra were extracted by summingflux inside large (18 px) apertures.

Visual inspection of the derived spectra quickly revealedresidual wavelength offsets of up to 0.5 nm, between target andreference stars, and wavelength drifts up to 0.3 nm, affecting thespectra of individual stars throughout each observing sequence.As our smallest wavelength bin is 2 nm, these drifts are largeenough to affect our results and we thus refined the wavelengthsolution on the target and reference spectra as follows beforeproceeding. To calculate a correct wavelength solution for allimages, we first removed telluric features with the molecfit rou-

600 700 800 900Wavelength [nm]

0

10

20

30

Flu

x [10

4 e

lectr

ons]

Fig. 1: Average spectra of target (top, black) and two referencestars. The 10 nm bins used for spectrophotometric extraction areindicated with blue dashed lines, while the purple dash-dottedlines indicate 5 nm bins centered on the Na and K features. Gapsbetween the GMOS detectors are shaded in gray.

tines (Smette et al. 2015, Kausch et al. 2015). We then calcu-lated improved wavelength solutions based on these correctedspectra and applied these to the individual (uncorrected) spectrabefore extracting spectrophotometry. The improved wavelengthsolutions were found in a three-step process: First, we identifiedthe best frames of each sequence as reference images and, foreach star, shifted all other spectra obtained during the same nightto the wavelength solution of these reference spectra. We did thisby cross-correlating a set of reference regions and, based on theinferred wavelength shifts, adapted the wavelength solution ofeach of the three detectors separately. Once all spectra of eachsequence and star were aligned to the respective reference spec-trum, we proceeded to finding an improved wavelength solution.To do so, we first combined all spectra of each star and date, pro-ducing a higher S/N spectrum. We then refined the wavelengthsolution of the target spectrum as follows. We fitted the contin-uum, normalized the target spectrum and cross-correlated a set ofwell-resolved features against a PHOENIX (Husser et al. 2013)model spectrum of an Teff=6100 K, log g=4.50 solar-metallicitystar that had been normalized and convolved down to the reso-lution of our data. Again, the wavelength solution was correctedby fitting cross-correlating a set of reference regions. Followingthe same procedure, we finally matched the two reference starspectra to that of the target.

Once we were certain of a reliable wavelength calibration,we binned the spectra of target and reference stars in 10 and20 nm wide wavelength bins and created relative transit lightcurves for each bin. To reach a higher wavelength resolution onthe Na and K features, we add 2, 5 and 10 nm bins centeredon these features. We find that the most precise light curves areobtained when only the fainter reference star is used instead ofthe the sum of both references, since the brighter reference starshowed strong correlated noise. See Figs. 3 and 4 for these spec-trophotometric light curves, together with the best models de-scribed below.

Article number, page 2 of 12

M. Lendl et al.: Signs of strong Na and K absorption in the transmission spectrum of WASP-103b

Wavelength [nm]

Flux

0.85

0.90

0.95

1.00

1.05

1.10

588.0 588.5 589.0 589.5 590.0 590.5

0.85

0.90

0.95

1.00

1.05

1.10

Fig. 2: Enlargement of the Na feature at 589 nm showing allspectra obtained on 07 June 2015. Flux is given in arbitraryunits, and the spectra are color-coded and shifted on the y-axisfor illustrative purposes. The top panel shows the drift of thewavelength solution throughout the sequence before wavelengthrecalibration, and the bottom panel shows the same sequence ofspectra after recalibration. Vertical blue dashed lines indicate thecenter and limits of our smallest wavelength bin around the Nafeature.

3. Lightcurve analysis

3.1. Fitting method

Our code uses a differential evolution Markov chain MonteCarlo (DEMCMC) approach for the combined analysis of transitlightcurves, combining the publicly available parallelized MC3code (Cubillos et al. 2017)1 with the prescription for transitlightcurve shape by Mandel & Agol (2002). Depending on theavailable resources (differential evolution MCMC methods relyon a large number of chains run in parallel), the sampling algo-rithm may be set to either a Metropolis-Hastings algorithm (e.g.Carlin & Louis 2009), a differential-evolution MCMC (DEM-CMC) (ter Braak 2006) algorithm, or the DEMC-z algorithmwith snooker proposals (ter Braak & Vrugt 2008). Convergenceis checked via the Gelman-Rubin (Gelman & Rubin 1992) test.

To account for photometric variations stemming from instru-mental, atmospheric or stellar effects, our code includes photo-metric baseline models, parametrizations of external variablessuch as time, stellar FWHM, and sky background. The paramet-ric baseline coefficients can either be included as jump parame-ters or calculated via least-square minimization at each MCMCstep (following Gillon et al. 2010). The later option can be ad-vantageous for computational reasons as it reduces the num-ber of jump parameters, however the first option allows to vi-sualize any correlations between photometric baseline parame-ters and inferred transit parameters. We assume a minimal base-line model assuming a second-order time polynomial, and ac-

1 https://github.com/pcubillos/MCcubed

cept more complicated models only if the Bayes factor (e.g.,Schwarz 1978) estimated from the Bayesian Information Cri-terion indicated significantly higher probability. It is also pos-sible to include a common prescription of correlated noise forspectrophotometric light curves from the same epoch and in-strument based on the common noise model (CNM) describedin Lendl et al. (2016). Before launching the MCMC, we run aleast-squares minimization and estimate the amplitudes of addi-tional white and red noise in each of the lightcurves via the βrand βw factors. βw is defined as the ratio of the residual RMSto the mean error and βr is calculated by comparing the resid-ual RMS of the binned to the residuals of the unbinned dataset (Winn et al. 2008, Gillon et al. 2010). For each lightcurve,we then rescale the errors by multiplying them with the productCF = βw×βw. This procedure also assures the correct weightingbetween several data sets of unequal quality.

The following parameters can be set as fitted (”jump”) pa-rameters in the analysis:

– the star to planet radius ratio Rp

R∗

– the impact parameter, b– the transit duration, T14

– the time of mid-transit, T0

– the orbital period, P–√

e sinω and√

e cosω, where e is the orbital eccentricity andω is the argument of periastron

– the linear combination of the quadratic limb-darkening coef-ficients (u1, u2) in each wavelength band, c1,i = 2u1,i + u2,iand c2,i = u1,i − 2u2,i (Holman et al. 2006)

– offsets(∆

Rp

R∗

)i

from a reference radius ratio(Rp

R∗

)0

for each

wavelength channel.(Rp

R∗

)0

is usually determined from broad-band photometry or “white”-light analysis binning all spec-trophotometric data together. It is not possible to fit

(∆

Rp

R∗

)i

and Rp

R∗simultaneously.

– the coefficients of the photometric baseline models

3.2. WASP-103 analysis

For the analysis of the GMOS WASP-103 data, we used theDEMCMC algorithm running between 30 and 180 parallel (de-pending on the number of fitted light curves) chains of 20000points each. To calculate limb darkening coefficients for all ourwavelength channels, we used the routines by Espinoza & Jordán(2015) together with the response function for the GMOS R400grism2. We calculated coefficients for a Teff = 6110.0 K, M/H= 0.00 and vturb = 2.0 km s−1 star for surface gravities oflog g = 4.0 and log g = 4.5 using interpolated PHOENIX mod-els (option P100). We then interpolated these values to match thesurface gravity of WASP-103, log g = 4.22. We kept the limb-darkening values fixed during our analysis. We verified that thisdid not impact our results by performing an additional run whileletting the limb darkening parameters vary, assuming only a wide(width 0.1, much larger than the variation of the parameters be-tween neighboring wavelength bins) Gaussian prior centered onthe model values. Results from this test were near-identical tothose of with fixed limb-darkening parameters.

2 available from https://www.gemini.edu/sciops/instruments/gmos/spectroscopy-overview/gratings

Article number, page 3 of 12

A&A proofs: manuscript no. Wasp-103_v5

200.90 200.95 201.00BJD -2456000 [d]

0.40

0.60

0.80

1.00

Rela

tive flu

x

564.10

584.10

604.10

624.10

644.10

661.53

681.30

705.00

725.00

745.00

765.00

785.00

803.60

824.40

848.30

868.30

888.30

908.30

928.30

948.05

589.30

589.30

589.30

768.20

768.20

768.20

766.46

769.88

213.85 213.90 213.95 214.00BJD -2456000 [d]

511.90 511.95 512.00BJD -2456000 [d]

Fig. 3: Spectrophotometric light curves of the three observation epochs (left, center, and right panels) obtained using 20 nm bins, aswell as bins centered on the Na and K line features. The central wavelength of each bin (in nm) is indicated in the left panel. Thelight curves centerend on the alkali features are ordered by bin width (from top to bottom): 10, 5 and 2 nm. Residuals are shownbelow, in the same order as the light curves.

Article number, page 4 of 12

M. Lendl et al.: Signs of strong Na and K absorption in the transmission spectrum of WASP-103b

200.90 200.95 201.00BJD -2456000 [d]

0.40

0.60

0.80

1.00

Rela

tive flu

x

559.10

569.10

579.10

589.10

599.10

609.10

619.10

629.10

639.10

649.10

659.10

666.53

677.15

685.45

700.00

710.00

720.00

730.00

740.00

750.00

760.00

770.00

780.00

790.00

800.00

808.60

820.20

828.60

843.30

853.30

863.30

873.30

883.30

893.30

903.30

913.30

923.30

933.30

943.30

953.05

213.85 213.90 213.95 214.00BJD -2456000 [d]

511.90 511.95 512.00BJD -2456000 [d]

Fig. 4: Spectrophotometric light curves of the three observation epochs (left, center, and right panels) obtained using 10 nm bins.The central wavelength of each bin (in nm) is indicated in the left panel. Residuals are shown below, in the same order as the lightcurves.

Article number, page 5 of 12

A&A proofs: manuscript no. Wasp-103_v5

3.2.1. Contamination

WASP-103 was identified to possess a nearby (0.24”) star byWöllert & Brandner (2015), which is blended with WASP-103in our data. Further observations by Ngo et al. (2016) and Cartieret al. (2017) reveal this contaminant to be a T = 4400 ±200 K K5V star, most likely physically associated to WASP-103(Cartier et al. 2017). Third light, if unaccounted for, introducessystematic effects in transmission spectra which may be misin-terpreted as effects stemming from the planetary atmosphere. Tocorrect our measurements, we estimate the contaminating flux inthe observed wavelength bands using PHOENIX model spectra(Husser et al. 2013) and the target and contaminant properties ofCartier et al. (2017). We binned PHOENIX model spectra to thespectral bins of our transmission spectrum and interpolated themto the target and contaminant parameters. We then calculated thecontaminating flux ratio for each wavelength bin i as(

Fcont

FW103

)i=

(Rcont

RW103

)2 (Mcont

MW103

)i, (1)

where (MW103,Mcont)i are the binned model spectra, andRcont/RW103 is the contaminant/target radius ratio. To estimatethe uncertainties, we drew 10000 temperature values from Gaus-sian distributions of target and contaminant values centered onthe values given by Cartier et al. (2017), and with a standard de-viation of their respective 1 − σ errors. As the radius ratio is notindependent of the stellar temperature, it needs to be estimatedfor each combination of target and contaminant temperatures.We did this by interpolating and binning model spectra for alltemperature values and using these together with the measuredJ, H and K band magnitude ratios of (Cartier et al. 2017) to in-fer appropriate radius ratios for our sample of target and con-taminant temperatures. For this set of 10000 simulated pairs ofspectra and radius ratios, we inferred the resulting flux ratios inthe spectral bins of our analysis and estimated the errors on theassumed flux ratios from this distribution. The obtained valuesand uncertainties are given in Table 1. These values are in goodagreement with those given by Southworth & Evans (2016) andCartier et al. (2017). We note that errors in the assumed contam-ination do not affect each spectral measurement independently,but rather cause horizontal shifts of the full transmission spec-trum, and/or introduce a wavelength-dependent slope. To evalu-ate any such effect on our transmission spectrum, we derive theimpact of each element of our sample on the measured RP/R∗values and then infer the 3-σ contours. These contours are shownas dotted lines in Fig. 7, and indicate that, while the measure-ments may be shifted up or down by 0.0013 in Rp/R∗, potentiallyintroduced slopes are small.

3.2.2. White light curves

We first performed an independent analysis of each of the whitelight curves obtained by combining the full spectra of each date.Here, we tested a wide variety of parametric baseline modelsto find the optimal fit to the observations of each date, select-ing a more complicated model over a simple one only if theBayes factor indicates significantly higher probability. Error barswere not adapted to compensate for red or white noise at thisstep, as this would bias the inferred BIC values. The observa-tions of 27 June 2015 and 10 July 2015 are best fit by poly-nomials of second order in time together with a sine function,f (t) = A0 + A1t + A2t2 + B0 sin(B1t + B2) (a common behav-ior for GMOS data, see Stevenson et al. 2014, von Essen et al.

2017), while for the light curve of 03 May 2016, a second or-der time polynomial is adequate. Using these baseline models,and allowing for error scaling to compensate for excess noise,we infer Rp/R∗ values of 0.1120 ± 0.0021, 0.1158 ± 0.0011, and0.1157 ± 0.0007 for the light curves observed on 27 June 2015,10 July 2015, and 03 May 2016, respectively. These values arein good agreement, with the two latter light curves producingnear-identical values and the first light curve, which is also theleast precise, showing a marginally smaller value. This is in linewith the non-detection of rotational modulation (which wouldlead to time-variable transit depths) for the host star (Gillon et al.2014). From a combined fit of all three white light curves, we de-rive the transit parameters listed in Table 2. Again, these valuesare in good agreement (< 1.5σ difference) with those publishedby Southworth et al. (2015), Southworth & Evans (2016) andCartier et al. (2017).

3.2.3. Inference of the transmission spectrum

We extracted spectrophotometry covering the full spectral rangefor wavelength intervals of 10 nm and 20 nm width, as well asfor 2 nm, 5 nm and 10 nm windows centered on the Na fea-ture. Due to the separation of the K feature components (whichis 3.4 nm, compared to 0.6 nm for the Na feature), we placedtwo 2 nm bins at the center of the two components and one binbetween them, in addition to 5 nm and 10 nm bins covering thewhole feature. We analyzed the full-range 10 and 20-nm bins asseparate sets, each time including the narrow bins centered onNa and K. Instead of fixing the transit shape parameters (T0, b,and T14) when inferring the transmission spectrum, which mayresult in under-estimated errors, we let these parameters vary,but imposed normal priors centered on the inferred value and1-σ errors obtained from the analysis of the white light curves.We used the white Rp/R∗ values measured above as referenceradii used in the calculation of the common noise model and theplanetary radius variations (∆Rp/R∗)i.

We first carried out individual analyses of the transmissionspectra obtained at the individual epochs, to search for any vari-ations. For each set, we used the best parametric model foundfrom the analysis of the white light curves (see Sect. 3.2.2), fit-ting all coefficients of the time polynomial as well as the ampli-tude of the sine function. We also tested the addition of a com-mon noise model, finding that its addition is warranted by signif-icant BIC improvements for all three dates. Additionally, the re-sulting transmission spectra obtained when including a commonnoise model show better agreement between the three epochs.We find that the spectra obtained on the three dates are in rea-sonable overall agreement with one another (see the upper panelof Fig. 5), with the discrepancy of two measurements in the samewavelength bin at a level of 0.65 σ and below 2.3 σ in all cases.For the 10 nm bins, agreement is at a similar level, with a medianof 0.55 σ and individual measurements disagreeing by as muchas 2.5 σ. The differences are most pronounced at short or longwavelengths, which could indicate that limited signal to noisecombined with larger supplementary wavelength corrections aslimiting factors.

To find the optimal transmission spectrum based on all avail-able data, we performed a combined analysis of all GMOS data,allowing for a single planetary radius offset (∆Rp/R∗)i per wave-length channel. The resulting transmission spectra are shown inFigs. 5 (20 nm bins) and 6 (10 nm bins), and given in Table 3.

Article number, page 6 of 12

M. Lendl et al.: Signs of strong Na and K absorption in the transmission spectrum of WASP-103b

Table 1

Wavelength bins, contamination flux ratios (Fcont/FW103) for WASP-103 and the nearby source, and quadratic limb-darkeningcoefficients used in the analysis.

Wavlength range [nm] Fcont/FW103 u1 u2 Wavelength range [nm] Fcont/FW103 u1 u2

10 nm bins 20 nm bins554.1 – 564.1 0.047 ± 0.0099 0.499 0.208 554.1 – 574.1 0.049 ± 0.0098 0.493 0.209564.1 – 574.1 0.051 ± 0.0098 0.488 0.210 574.1 – 594.1 0.051 ± 0.0099 0.472 0.212574.1 – 584.1 0.053 ± 0.0097 0.477 0.211 594.1 – 614.1 0.056 ± 0.0097 0.450 0.217584.1 – 594.1 0.049 ± 0.0103 0.466 0.213 614.1 – 634.1 0.056 ± 0.0100 0.432 0.219594.1 – 604.1 0.055 ± 0.0099 0.459 0.212 634.1 – 654.1 0.060 ± 0.0097 0.399 0.232604.1 – 614.1 0.056 ± 0.0095 0.449 0.214 654.1 – 668.9 0.065 ± 0.0094 0.383 0.235614.1 – 624.1 0.056 ± 0.0101 0.436 0.218 673.0 – 689.6 0.063 ± 0.0108 0.383 0.226624.1 – 634.1 0.057 ± 0.0100 0.429 0.218 695.0 – 715.0 0.067 ± 0.0095 0.366 0.197634.1 – 644.1 0.059 ± 0.0100 0.426 0.187 715.0 – 735.0 0.070 ± 0.0089 0.350 0.199644.1 – 654.1 0.061 ± 0.0094 0.398 0.201 735.0 – 755.0 0.073 ± 0.0080 0.350 0.213654.1 – 664.1 0.065 ± 0.0093 0.384 0.236 755.0 – 775.0 0.075 ± 0.0082 0.339 0.214664.1 – 668.9 0.064 ± 0.0096 0.393 0.227 775.0 – 795.0 0.077 ± 0.0080 0.318 0.232673.0 – 681.3 0.063 ± 0.0106 0.387 0.226 795.0 – 812.2 0.079 ± 0.0074 0.308 0.234681.3 – 689.6 0.062 ± 0.0112 0.381 0.226 816.0 – 832.8 0.081 ± 0.0072 0.292 0.236695.0 – 705.0 0.066 ± 0.0094 0.371 0.227 838.3 – 858.3 0.083 ± 0.0072 0.276 0.242705.0 – 715.0 0.067 ± 0.0097 0.361 0.198 858.3 – 878.3 0.086 ± 0.0070 0.278 0.237715.0 – 725.0 0.069 ± 0.0094 0.354 0.198 878.3 – 898.3 0.087 ± 0.0070 0.279 0.232725.0 – 735.0 0.071 ± 0.0085 0.350 0.230 898.3 – 918.3 0.089 ± 0.0067 0.275 0.231735.0 – 745.0 0.072 ± 0.0081 0.343 0.230 918.3 – 938.3 0.091 ± 0.0062 0.273 0.228745.0 – 755.0 0.073 ± 0.0080 0.347 0.213 938.3 – 957.8 0.093 ± 0.0061 0.261 0.232755.0 – 765.0 0.075 ± 0.0078 0.348 0.206765.0 – 775.0 0.075 ± 0.0084 0.331 0.225 Na feature775.0 – 785.0 0.077 ± 0.0082 0.320 0.232 584.3 – 594.3 0.049 ± 0.010 0.465 0.215785.0 – 795.0 0.078 ± 0.0077 0.314 0.233 586.8 – 591.8 0.044 ± 0.011 0.479 0.200795.0 – 805.0 0.079 ± 0.0075 0.314 0.224 588.3 – 590.3 0.032 ± 0.010 0.466 0.214805.0 – 812.2 0.079 ± 0.0073 0.310 0.225816.0 – 824.4 0.080 ± 0.0073 0.296 0.236 K feature824.4 – 832.8 0.082 ± 0.0072 0.288 0.239 763.2 – 773.2 0.075 ± 0.0084 0.328 0.230838.3 – 848.3 0.083 ± 0.0073 0.281 0.234 765.7 – 770.7 0.074 ± 0.0085 0.329 0.220848.3 – 858.3 0.084 ± 0.0072 0.280 0.234 767.2 – 769.2 0.076 ± 0.0084 0.325 0.231858.3 – 868.3 0.086 ± 0.0070 0.270 0.241 765.46 – 767.46 0.072 ± 0.0084 0.327 0.232868.3 – 878.3 0.086 ± 0.0071 0.273 0.236 768.88 – 770.8.8 0.074 ± 0.0086 0.326 0.232878.3 – 888.3 0.087 ± 0.0069 0.268 0.239888.3 – 898.3 0.088 ± 0.0070 0.265 0.240 white898.3 – 908.3 0.090 ± 0.0069 0.280 0.225 5541 – 9578 0.069 ± 0.0087 0.341 0.226908.3 – 918.3 0.089 ± 0.0066 0.279 0.223918.3 – 928.3 0.092 ± 0.0062 0.275 0.225928.3 – 938.3 0.091 ± 0.0064 0.273 0.223938.3 – 948.3 0.092 ± 0.0061 0.265 0.207948.3 – 957.8 0.095 ± 0.0061 0.254 0.212

Table 2

The transit parameters inferred from an analysis of the whitelight curves.

Parameter ValueTime of mid-transit, T0 [HJDUTC] 7511.943617 ± 0.000076Star/planet radius ratio, RP/R∗ 0.11442+0.00050

−0.00043Impact parameter, b 0.074+0.066

−0.048Transit duration, T14 [d] 0.11232 ± 0.00019Period [d] 0.92554705 ± 0.0000004

4. Results and Discussion

4.1. Model atmospheres

We compute model transmission spectra of WASP-103b, usingsolar abundances, to aid the interpretation of our observations.To produce atmospheric transmission forward models of WASP-103b, we use the open-source Python Radiative Transfer in aBayesian framework (Pyrat Bay) package3 (Cubillos et al. 2017,in prep.), which is based on the Bayesian Atmospheric Radia-tive Transfer package (Blecic 2016, Cubillos 2016). The modelsof WASP-103b consider molecular opacities for H2O (HITEMP,Rothman et al. 2010), collision induced absorption from H2-H2(Borysow et al. 2001, Borysow 2002) and H2-He (Borysow et al.

3 http://pcubillos.github.io/pyratbay

Article number, page 7 of 12

A&A proofs: manuscript no. Wasp-103_v5

-0.015

-0.010

-0.005

0.000

0.005

0.010

∆R

P /

RS

600 700 800 900Wavelength [nm]

-0.010

-0.005

0.000

0.005

0.010

∆R

P /

RS

Fig. 5: Inferred Rp/R∗ offsets for wavelength bins of using 20 nmwidth. Top: results from separate analyses to each epoch, whereblue, red and green points refer to results from 27 June 2015, 10July 2015, and 03 May 2016, respectively. Bottom: results fromthe combined analysis of all data (black points), together withthe 2 nm bins centered on the Na and K features. Results fromthe separate analyses are shown in gray.

-0.015

-0.010

-0.005

0.000

0.005

0.010

∆R

P /

RS

600 700 800 900Wavelength [nm]

-0.010

-0.005

0.000

0.005

0.010

∆R

P /

RS

Fig. 6: Inferred Rp/R∗ offsets for wavelength bins of using 20 nmwidth. Top: results from separate analyses to each epoch, whereblue, red and green points refer to results from 27 June 2015, 10July 2015, and 03 May 2016, respectively. Bottom: results fromthe combined analysis of all data (black points), together withthe 2 nm bins centered on the Na and K features. Results fromthe separate analyses are shown in gray.

1988, 1989, Borysow & Frommhold 1989), H2 Rayleigh scat-tering (Lecavelier Des Etangs et al. 2008b), and resonant alkalilines (Burrows et al. 2000).

The code computes line-by-line radiative-transfer for a 1Datmospheric model consisting of concentric shell layers, in hy-drostatic equilibrium, and thermochemical-equilibrium abun-dances (Blecic et al. 2016). We assumed isothermal temper-ature profiles with a temperature of 2500 K, and calculatedthermochemical-equilibrium models from solar elemental abun-dances scaling the metallicities to 0.1, 1, and 10 times the solarvalue.

4.2. Spectral slope by scattering or contamination

We use our combined transmission spectra to study the atmo-spheric properties of WASP-103b. Southworth et al. (2015) usedbroadband light curves to infer the presence of a strong scatteringslope in the transmission spectrum of WASP-103b, which holds

even after correcting their measurements from contamination bythe nearby star (Southworth & Evans 2016). While our data donot span the full extent in wavelength of their measurements, inparticular we have no data bluewards of 560 nm, we do not ob-serve any evident trend in the transmission spectrum. To test this,we fitted a straight line with a linear model in wavelength to the10 nm bin data set, excluding the narrow wavlength bins cen-tered on the Na and K features. We found a negligible slope ofabsorption increasing with wavelength 1.04× 10−6 ± 1.54× 10−6

in RP/R∗ per nm (best values and errors were determined via anMCMC bootstrap approach drawing 10000 subsamples). Com-paring this solution with a straight horizontal line, we find noevidence of the slope producing a significantly more accurate fitto the data (Bayes factor 0.2). We show our measurements nextto those of Southworth & Evans (2016) in Fig. 7. Next to thedata, we show our best-fit slope as well as a Rayleigh scatter-ing slope inferred from the broadband data as in Southworth &Evans (2016). The latter is parametrized, following LecavelierDes Etangs et al. (2008a), by dRp/R∗(λ)

d ln λ = −0.0066(17). Addi-tional observations at shorter wavelengths would be helpful todefinitely measure any scattering signature in the atmosphere ofWASP-103b.

Beside the physical properties of the planetary atmosphere,blended third light from a companion star affects the transmis-sion spectrum of WASP-103b. While we correct for this thirdlight in our analysis, uncertainties in the assumed spectral typeof target and contaminating star can lead to residual slopes andoverall offsets. As described in Sect. 3.2.1, we have carried out alarge set of simulations, perturbing the assumed temperatures oftarget and contaminant within the errors given by Cartier et al.(2017). We find that even in the most extreme cases, the addedslopes are too small to significantly impact the shape of our ob-served transmission spectrum. However, our transmission spec-trum may be subject to an overall offset of ±0.0013 in RP/R∗.This is illustrated in Fig. 7, where two dotted gray lines indi-cate the potentially added slopes (assuming the 3-σ envelop toall simulated cases).

4.3. Na and K absorption features

Absorption by Na and K produces prominent features in plane-tary transmission spectra. As WASP-103b has a high mass andhence density, the predicted amplitude of these features is smallcompared to other planets studied previously with ground-basedtransmission spectroscopy (e.g., 2.7 times smaller than those ofthe recently-studied WASP-39b, Nikolov et al. 2016, and WASP-17b ,Sedaghati et al. 2016), these features are inherently difficultto detect for WASP-103b. Our data cover both the 590 nm Naand the 766 nm K doublets, and we place a set of narrow binson these features to test for extra absorption. As shown in Fig.8, we detect signs of added absorption at the line cores of bothfeatures. The Na feature is more clearly detected. Even thoughthe broadband transmission spectrum shows a trend (of unknownorigin) across the 560 – 630 nm region, the measured absorptionis enhanced for the 2 nm bin centered on the Na feature com-pared to the neighboring points (see Fig. 8). We resolve the twocomponents of the K feature by placing a 2 nm wide bin on eachof the two peaks as well as on 2 nm wide bin between them. Fur-ther, we place bins of 5 and 10 nm centrally on the feature. Bothcomponents show enhanced absorption compared to the neigh-boring spectral bins, or the spectral bin placed between the twocomponents.

Article number, page 8 of 12

M. Lendl et al.: Signs of strong Na and K absorption in the transmission spectrum of WASP-103b

1000 1200 1400 1600 1800 2000Wavelength [nm]

HST/WFC3

400 500 600 700 800 900Wavelength [nm]

0.105

0.110

0.115

0.120

RP

/ RS

Fig. 7: Optical transmission spectrum of WASP-103b. Measurements obtained with 10 and 20 nm wide bins are shown as brownfilled circles and black squares, respectively. Measurements in 2 nm wide bins centered on the Na feature, as well as on bothcomponents of the K feature are shown in magenta. Observations by Southworth & Evans (2016) are indicated by blue squares,together with their inferred spectral sloped (blue dashed line). The best-fit spectral slope inferred from our data is shown as a graydashed line. Model transmision spectra for WASP-103b’s are shown as a light green (0.1 × solar metallicity), dark green (solarmetallicity) and olive (10 × solar metallicity) lines and filled circles indicate the models binned to the resolution of the data. Below,the two dotted lines indicate the possible added spectral slopes (at 3-σ level) due to uncertainties in the properties of WASP-103band the contaminating star. The vertical scale is identical to that of the upper panel. In the right panel, the shaded region indicatesthe wavelength coverage of HST/WFC3’s G141 grism, and the expected precision of spectrophotometric WFC3 measurements isindicated.

For both features, the absorption appears to be slightly larger(0.74, 0.36 and 0.43 σ for the Na and the blue and red compo-nents of the K line, respectively) than predicted from our modelspectra. Enhanced absorption features could be related to theplanet’s tidally distorted state, for which we calculate (using theRoche lobe model by Budaj 2011, as well as stellar parametersfrom Southworth & Evans 2016) a substellar radius 10.5% largerthan the planet’s polar radius. More detailed simulations of theeffect of tidal distortion on transmission spectra are however be-yond the scope of this work.

As demonstrated by Deming & Sheppard (2017), observa-tions at low to medium spectral resolution are affected by an res-olution linked bias (RLB), which acts to decrease the apparentabsorption feature amplitude. This is due to flux leakage fromneighboring spectral regions into the absorption line, where theplanetary atmosphere’s enhanced absorption is located. We haveevaluated the RLB affecting our measurements of the Na andK lines following the formalism of Deming & Sheppard (2017)(Equations 3 – 5), and assuming an instrumental spectral resolu-tion of 1200 together with our smallest bin size of 2 nm. We findthat the effect is small compared to our uncertainties, 2.0× 10−4,5.4 × 10−5, and 2.9 × 10−5 in ∆Rp/R∗ for the 2 nm bins centeredon the Na and the blue and red components of the K feature,respectively.

4.4. Atmospheric metallicity and cloud altitude

We compare our data to model atmospheres calculated for metal-licities 0.1, 1 and 10 times the solar value. The data are mostclosely reproduced by models with super-solar metal abun-

dances, evidenced by χ2 values of 51.1, 56.0 and 60.0 whencomparing data to models of 10, 1, and 0.1 solar metallicity. Thisis due to slightly enhanced absorption measured at wavelengthslarger than 900 nm, where water absorption becomes prominent.Extending WASP-103’s transmission spectrum to longer wave-lengths, targeting prominent water absorption features, would berequired to make a more solid statement on the planet’s atmo-spheric composition. We show the predicted transmission spec-tra in the near-IR in the right panel of Fig. 7, indicating the lo-cation of HST’s WFC3 G141 grism, and precision estimate of50 ppm (e.g., Kreidberg et al. 2015). Water should be easily de-tectable in the atmosphere of WASP-103b.

We do not observe any evidence for aerosols in the planetarytransmission spectrum. This means that, if clouds or hazes arepresent in the atmosphere of WASP-103b, they must be locatedat altitudes below (or pressure levels larger than) those probedby our measurements. For the models calculated here, the atmo-spheric pressure levels probed outside the alkali features rangebetween 0.01 bar (for 10 × solar metallicity) and 0.1 bar (for 0.1× solar metallicity). If a cloud deck is present in the atmosphereof WASP-103b, it must reside at pressures higher than 0.01 bar.This upper limit is well inside the range of inferred cloud deckaltitudes for hot Jupiters (e.g., Barstow et al. 2017).

4.5. The atmosphere of WASP-103b in context

Based on these measurements, we tentatively classify WASP-103b as a planet with a clear atmosphere. The absence of awavelength-dependent slope in our data, combined with the de-tection of alkali absorption indicates that aerosols, if present in

Article number, page 9 of 12

A&A proofs: manuscript no. Wasp-103_v5

570 580 590 600 610Wavelength [nm]

0.110

0.115

0.120

RP / R

S

Na feature

760 770 780Wavelength [nm]

0.110

0.112

0.114

0.116

0.118

0.120

RP / R

S

K feature

Fig. 8: Transmission spectrum of WASP-103b zoomed in on theNa (top) and K (bottom) features. The black points refer to the10 nm wavelength bins also shown in Fig. 6, blue points indicate2 nm bins centered on the features: either one bin centered onthe Na feature, or two point centered on each component of theK feature. In the lower panel, brown refers to a 2 nm wide bincentered between the two components. Red and orange pointsdenote wavelength bins of 5 and 10 nm width, respectively, cen-tered on the features. The solar-composition model spectrum isshown in green, and filled circles indicate the model binned tothe resolution of the data.

substantial quantities in the atmosphere of WASP-103b, existlargely at altitudes below those probed by tranmsission spec-troscopy. The planet follows the trends of Stevenson (2016),who proposed that hot planets, and planets with high surfacegravities, are more likely to possess large water absorption fea-tures expected from clear atmospheres. Along the same lines, theplanet also follows the cloudiness-temperature relation proposedby Heng (2016), who find that alkali absorption features are mostpronounced for hot planets.

In terms of mass and temperature, WASP-103b greatly re-sembles WASP-12b, an object for which a detailed transmissionspectrum has been observed (Mandell et al. 2013, Sing et al.2013, Swain et al. 2013, Stevenson et al. 2014, Kreidberg et al.2015). At the precision of the available data, the optical trans-mission spectrum of WASP-12b does not show alkali absorptionfeatures, however it does show a wavelength-dependent slope,which can be interpreted as signs of an aerosol layer (Sing et al.2013, Barstow et al. 2017). Water has been detected (Kreidberget al. 2015) in near-IR observations, but also these observationsare best fit when including an additional atmospheric opacitysource. It appears thus that the spectrum of WASP-12b is morestrongly affected by clouds or hazes than that of WASP-103b.However, direct comparison between results for the two planets

is not straight-forward as our observations of WASP-103b have adifferent resolution and wavelength coverage than those by (Singet al. 2013) of WASP-12b. Our data have a higher spectral res-olution, and we are thus more sensitive to absorption in the al-kali line cores, which may have gone undetected for WASP-12b.At the same time, our data cover a more narrow wavelength-region and are thus less sensitive to scattering slopes. Based onour detection of alkali features in the optical part of the spec-trum, we would expect WASP-103b to show also a prominentwater feature in its near-IR transmission spectrum. Near-IR ob-servations, but also observations spanning a wider wavelengthregion in the optical will allow a more detailed characterizationof WASP-103b’s atmosphere. Also, higher-resolution optical ob-servations of WASP-12b would be an asset to make the availabledata on these two planets truly comparable.

5. Conclusions

We present an optical transmission spectrum of the highly irra-diated exoplanet WASP-103b using Gemini/GMOS covering thewavelength range between 550 and 960 nm. WASP-103b is oneof the closest-orbiting hot Jupiters known, with an equilibriumtemperature of 2500 K and an orbital separation of less than 1.2times the Roche limit. From a combined analysis of observationsfrom three individual transits, we find enhanced absorption at the589 nm Na and, at lesser significance, at the 766 nm K feature.This indicates that the transmission spectrum of WASP-103b isnot dominated by strong aerosol absorption, a finding that is inline with previously-published trends of aerosol appearance andplanetary properties (Heng 2016, Stevenson 2016). Based on thealtitudes probed by our data, we constrain the pressure at thetop of any potential clouds in the atmosphere of WASP-103 tobe at least 0.01 bar. At low significance, the Na feature as wellas both components of the K feature, show slightly enhancedabsorptions compared to predictions from a solar-compositionplanetary atmosphere model.

Our observations do not confirm a previously-inferred strongtrend for increasing absorption at small wavelengths (South-worth et al. 2015, Southworth & Evans 2016). We have studiedthe impact of contamination from a blended star on the transmis-sion spectrum, but fail to produce a trend large enough to explainprevious observations even when assuming errors on the stellarparameters of both stars at the 3-σ level.Acknowledgements. We thank an anonymous referee for insightful commentsthat improved the quality of this work. We acknowledge Laetitia Delrez andCatherine Huitson for valuable discussions on the treatment of contaminationand the wavelength stability of GMOS. IJ acknowledgements support fromthe Austrian Research Promotion Agency (FFG) under grant number P847963.Based on observations obtained at the Gemini Observatory, which is operated bythe Association of Universities for Research in Astronomy, Inc., under a cooper-ative agreement with the NSF on behalf of the Gemini partnership: the NationalScience Foundation (United States), the National Research Council (Canada),CONICYT (Chile), Ministerio de Ciencia, Tecnología e Innovación Productiva(Argentina), and Ministério da Ciência, Tecnologia e Inovação (Brazil).

ReferencesBarstow, J. K., Aigrain, S., Irwin, P. G. J., & Sing, D. K. 2017, ApJ, 834, 50Blecic, J. 2016, ArXiv e-prints [arXiv:1604.02692]Blecic, J., Harrington, J., & Bowman, M. O. 2016, ApJS, 225, 4Borysow, A. 2002, A&A, 390, 779Borysow, A. & Frommhold, L. 1989, ApJ, 341, 549Borysow, A., Frommhold, L., & Moraldi, M. 1989, ApJ, 336, 495Borysow, A., Jorgensen, U. G., & Fu, Y. 2001, J. Quant. Spectr. Rad. Transf., 68,

235Borysow, J., Frommhold, L., & Birnbaum, G. 1988, ApJ, 326, 509

Article number, page 10 of 12

M. Lendl et al.: Signs of strong Na and K absorption in the transmission spectrum of WASP-103b

Table 3

The transmission spectrum of WASP-103b as observed with GMOS.

Wavlength range [Å] (∆Rp/R∗)i Wavelength range [Å] (∆Rp/R∗)i

10 nm bins 20 nm bins5541 – 5641 −0.0035 ± 0.0024 5541 – 5741 −0.00356 ± −0.00145641 – 5741 −0.0024 ± 0.0019 5741 – 5941 −0.00137 ± −0.00125741 – 5841 −0.0014 ± 0.0019 5941 – 6141 0.00029 ± −0.00125841 – 5941 −0.0003 ± 0.0016 6141 – 6341 0.00299 ± −0.00105941 – 6041 0.0006 ± 0.0016 6341 – 6541 −0.00161 ± −0.000876041 – 6141 0.0004 ± 0.0016 6541 – 6689 −0.00334 ± −0.000976141 – 6241 0.0024 ± 0.0012 6730 – 6896 −0.00259 ± −0.000996241 – 6341 0.0035 ± 0.0013 6950 – 7150 −0.00101 ± −0.000806341 – 6441 −0.0009 ± 0.0012 7150 – 7350 0.00124 ± −0.000686441 – 6541 −0.0019 ± 0.0012 7350 – 7550 0.00047 ± −0.000666541 – 6641 −0.0049 ± 0.0012 7550 – 7750 0.00087 ± −0.000956641 – 6689 −0.0007 ± 0.0015 7750 – 7950 −0.00140 ± −0.000576730 – 6813 −0.0028 ± 0.0012 7950 – 8122 −0.00051 ± −0.000806813 – 6896 −0.0024 ± 0.0012 8160 – 8328 −0.00192 ± −0.000886950 – 7050 −0.0021 ± 0.0012 8383 – 8583 −0.00064 ± −0.000827050 – 7150 0.00008 ± 0.00086 8583 – 8783 0.00033 ± −0.000837150 – 7250 0.00195 ± 0.00095 8783 – 8983 −0.0003 ± −0.00107250 – 7350 0.00086 ± 0.00088 8983 – 9183 0.00018 ± −0.000837350 – 7450 −0.00009 ± 0.00090 9183 – 9383 0.0005 ± −0.00127450 – 7550 0.00081 ± 0.00088 9383 – 9578 0.0004 ± −0.00137550 – 7650 −0.0007 ± 0.00197650 – 7750 0.00151 ± 0.00083 Na feature7750 – 7850 −0.00111 ± 0.00080 5843 – 5943 0.0003 ± −0.00167850 – 7950 −0.00146 ± 0.00089 5868 – 5918 −0.0021 ± −0.00237950 – 8050 0.0010 ± 0.0011 5883 – 5903 0.0061 ± −0.00398050 – 8122 −0.0019 ± 0.00128160 – 8244 −0.0023 ± 0.0010 K feature8244 – 8328 −0.0016 ± 0.0013 7632 – 7732 0.0011 ± −0.00108383 – 8483 0.0012 ± 0.0010 7657 – 7707 0.0018 ± −0.00128483 – 8583 −0.00316 ± 0.00098 7672 – 7692 −0.0018 ± −0.00208583 – 8683 0.0012 ± 0.0010 7654.6 – 7674.6 0.0030 ± −0.00218683 – 8783 0.0005 ± 0.0012 7688.8 – 7708.8 0.0028 ± −0.00208783 – 8883 0.0014 ± 0.00148883 – 8983 −0.0014 ± 0.00118983 – 9083 −0.0003 ± 0.00129083 – 9183 −0.0008 ± 0.00159183 – 9283 0.0000 ± 0.00159283 – 9383 0.0019 ± 0.00229383 – 9483 −0.0011 ± 0.00189483 – 9578 0.0035 ± 0.0021

Budaj, J. 2011, AJ, 141, 59Burrows, A., Marley, M. S., & Sharp, C. M. 2000, ApJ, 531, 438Carlin, B. & Louis, T. 2009, Bayesian methods for data analysis, Texts in statis-

tical science (CRC Press)Cartier, K. M. S., Beatty, T. G., Zhao, M., et al. 2017, AJ, 153, 34Charbonneau, D., Brown, T. M., Noyes, R. W., & Gilliland, R. L. 2002, ApJ,

568, 377Chen, G., Palle, E., Nortmann, L., et al. 2017, ArXiv e-prints

[arXiv:1703.06716]Cubillos, P., Harrington, J., Loredo, T. J., et al. 2017, AJ, 153, 3Cubillos, P. E. 2016, ArXiv e-prints [arXiv:1604.01320]Deming, D. & Sheppard, K. 2017, ApJ, 841, L3Deming, D., Wilkins, A., McCullough, P., et al. 2013, ApJ, 774, 95Espinoza, N. & Jordán, A. 2015, MNRAS, 450, 1879Gelman, A. & Rubin, D. 1992, Statist. Sci., 7, 457Gillon, M., Anderson, D. R., Collier-Cameron, A., et al. 2014, ArXiv e-prints

[arXiv:1401.2784]Gillon, M., Lanotte, A. A., Barman, T., et al. 2010, A&A, 511, A3Heng, K. 2016, ApJ, 826, L16Holman, M. J., Winn, J. N., Latham, D. W., et al. 2006, ApJ, 652, 1715

Husser, T.-O., Wende-von Berg, S., Dreizler, S., et al. 2013, A&A, 553, A6Kausch, W., Noll, S., Smette, A., et al. 2015, A&A, 576, A78Kreidberg, L., Line, M. R., Bean, J. L., et al. 2015, ApJ, 814, 66Lammer, H., Selsis, F., Ribas, I., et al. 2003, ApJ, 598, L121Lecavelier Des Etangs, A., Pont, F., Vidal-Madjar, A., & Sing, D. 2008a, A&A,

481, L83Lecavelier Des Etangs, A., Vidal-Madjar, A., Désert, J.-M., & Sing, D. 2008b,

A&A, 485, 865Leconte, J., Lai, D., & Chabrier, G. 2011, A&A, 528, A41Lendl, M., Delrez, L., Gillon, M., et al. 2016, A&A, 587, A67Mandel, K. & Agol, E. 2002, ApJ, 580, L171Mandell, A., Haynes, K., Sinukoff, E., et al. 2013, ArXiv e-prints

[arXiv:1310.2949]Matsumura, S., Peale, S. J., & Rasio, F. A. 2010, ApJ, 725, 1995Ngo, H., Knutson, H. A., Hinkley, S., et al. 2016, ApJ, 827, 8Nikolov, N., Sing, D. K., Gibson, N. P., et al. 2016, ApJ, 832, 191Patra, K. C., Winn, J. N., Holman, M. J., et al. 2017, AJ, 154, 4Pont, F., Tamuz, O., Udalski, A., et al. 2008, A&A, 487, 749Rothman, L. S., Gordon, I. E., Barber, R. J., et al. 2010,

J. Quant. Spectr. Rad. Transf., 111, 2139

Article number, page 11 of 12

A&A proofs: manuscript no. Wasp-103_v5

Schwarz. 1978, Annals of Statistics, 6, 461Sedaghati, E., Boffin, H. M. J., Jerabková, T., et al. 2016, A&A, 596, A47Sing, D. K., Fortney, J. J., Nikolov, N., et al. 2016, Nature, 529, 59Sing, D. K., Lecavelier des Etangs, A., Fortney, J. J., et al. 2013, MNRAS, 436,

2956Sing, D. K., Wakeford, H. R., Showman, A. P., et al. 2015, MNRAS, 446, 2428Smette, A., Sana, H., Noll, S., et al. 2015, A&A, 576, A77Southworth, J. & Evans, D. F. 2016, MNRAS, 463, 37Southworth, J., Mancini, L., Ciceri, S., et al. 2015, MNRAS, 447, 711Stevenson, K. B. 2016, ApJ, 817, L16Stevenson, K. B., Bean, J. L., Seifahrt, A., et al. 2014, AJ, 147, 161Swain, M., Deroo, P., Tinetti, G., et al. 2013, Icarus, 225, 432ter Braak, C. J. F. & Vrugt, J. A. 2008, Statistics and Computing, 18, 435ter Braak, C. J. F. T. 2006, Statistics and Computing, 16, 239Udry, S., Mayor, M., & Santos, N. C. 2003, A&A, 407, 369van Dokkum, P. G. 2001, PASP, 113, 1420Vidal-Madjar, A., Lecavelier des Etangs, A., Désert, J.-M., et al. 2003, Nature,

422, 143von Essen, C., Cellone, S., Mallonn, M., et al. 2017, ArXiv e-prints

[arXiv:1703.10647]Winn, J. N., Holman, M. J., Torres, G., et al. 2008, ApJ, 683, 1076Wöllert, M. & Brandner, W. 2015, A&A, 579, A129

Article number, page 12 of 12